Why your next CPU could be a GPU

How GPUs could bring supercomputing to the desktop

Everybody's talking about supercomputing on the desktop – and in particular, whether it will be GPUs that achieve that goal. We think that general-purpose computation on GPUs (an idea known as the GPGPU) might be the most important computing trend over the next 10 years.
As claims go, it's a biggie. But if you want proof of the industry's faith in the new concept, just take a look at the companies that want a slice of the GPGPU pie: Nvidia, AMD, Intel, Microsoft, IBM, Apple and Toshiba all want in. And it's not just speculation that's leading to such big interest: GPGPU systems are already outperforming CPU-only clusters in fields as diverse as molecular dynamics, ray tracing, medical imaging and sequence matching.
The combination of parallel CPU and GPU processing used to achieve these results is often dubbed 'heterogeneous computing'. The GPGPU concept enables the GPU to moonlight as a versatile co-processor. As Nvidia's David Luebke has suggested, computers are no longer getting faster; the move to multicore processors means that they're actually getting wider.
That's the idea that GPGPU computing cashes in on. By intelligently offloading data intensive tasks from the CPU to other processor cores (such as those in a graphics card), developers achieve improved application performance through parallelism.
The GPGPU is hardly a new idea, however. According to website www.gpgpu.org, GPU technology has been used for number crunching since 1978, when Ikonas developed a programmable raster display system for cockpit instrumentation.
From GPU to GPGPU
Modern GPUs make ideal co-processors. Not only are they cheap, they're also blisteringly fast, thanks to the presence of multiple processor cores. Most importantly, these multiple cores are programmable. While CPUs are designed to process threads sequentially, GPUs are designed to burn through data in parallel.
The Nvidia GeForce GTX 280, for example, is built for speed. As a gaming component, it's capable of delivering smooth high-definition visuals with complex lighting effects, textures and realtime physics. Just take a look at Far Cry 2 in 1,920 x 1,200 pixels. With 1.4 billion transistors, the GeForce GTX 280 commands 240 programmable shader cores that can provide 993 gigalops of processing power.
AMD's graphics technology is equally potent. Its 4800 Series Radeon HD cards feature 800 programmable cores and GDDR5 memory to deliver 1.2 teralops of processing power. "Strict pipelining of GPU programs enables efficient access to data," says Shankar Krishnan at AT&T's Research Labs. "This obviates the need for the extensive cache architectures needed on traditional CPUs and allows for a much higher density of computational units."
Of course, if you're not playing Far Cry 2 or Fallout 3 then all this processing potential is just sitting about twiddling its thumbs. GPGPUs will allow us to see what will happen if other applications are able to make use of the processors in a graphics card.
Stream processing
This is why Nvidia and AMD are keen to harness the GPGPU potential of their graphics hardware. Nvidia's Tesla Personal Supercomputer, for example, combines a traditional quad-core workstation CPU with three or four Tesla C1060 processors.
A C1060 is effectively a GeForce GTX 280 with 4GB of GDDR3 memory and no video-out. Each C1060 is capable of 933 gigalops of single-precision floating point performance, so Nvidia's top-of-the range four-GPU S1070 system packs up to 4.14 teralops of processing power in each rack. The Tokyo Institute of Technology recently bought 170 of them to give its Tsubame supercomputer some extra kick.

Intel Nehalem Xeon X5570 2.93GHz review

Intel's new Nehalem CPU architecture is staggeringly powerful

Around 2004, it became clear that Intel's quest for clockspeed had burned out. The final "Prescott" revision of the Pentium 4 processor architecture was struggling to meet ambitious targets for operating frequency. A new paradigm in CPU design was needed.
Since then, Intel has been working to re-architect its entire family of x86 processors. First came the Core Duo mobile processor. It revolutionised notebook performance and gave an early glimpse of the enormous performance-per-watt pay off possible with a relatively low-clocking multi-core processor. Next was the Core 2 Duo processor, which did much the same for desktop computing.
The final piece of the puzzle
This month, Intel is putting the final piece of its multi-core, high efficiency strategy into place. Its latest Nehalem processor architecture is now available in Xeon trim, enabling Intel to make a strong claim to best-in-class across all segments – mobile, desktop and enterprise.
Of course, the Nehalem architecture has already been available for several months in the shape of Intel's enormously powerful Core i7 desktop processor. However, the upgrading of the Xeon workstation and server family of processors to Nehalem specification is much more significant than Core i7. Partly, that's because Intel had a healthy performance advantage on the desktop. The arrival of Core i7 merely served to cement what was already an extremely solid lead.
But it's also because, more than anything, Nehalem has been designed to be an enterprise class chip with killer parallel processing performance. Central to that aim is the concept of scalability in multi-core configurations and multi-socket scenarios. Adding extra CPUs and cores to a given platform is all very well. But if those CPUs and cores are not fed with sufficient data, the consequence is idle processing cycles and compromised performance.
Until Nehalem, that's exactly the problem that Intel's multi-socket platforms have suffered from. Thanks to an old fashioned interface composed of a front side bus connecting the CPU to a secondary chip, which in turn housed the memory controller, system I/O and inter-socket communications, Intel's multi-socket Xeons were often starved of data. This shared bus was simply too slow and narrow to keep up with the demands of the CPUs, particularly as per-socket core counts increased.
Improved memory controller
But with Nehalem, Intel has moved both the memory controller and I/O onto the CPU itself. The result is a massive leap in available bandwidth. Take memory performance. Intel's outgoing dual-socket Xeon processor based on the Penryn architecture had to make do with at best 6GB/s of memory bandwidth for each CPU.
The new Xeon EPs based on Nehalem enjoy a staggering 16GB/s per socket. Of course, the boost in bandwidth is not only down to the controller's on-die location. Intel has also added an extra channel for a grand total of three and upped memory support to 1,333MHz DDR3. It's by far the most powerful memory controller of any x86 processor.Just as significant is the new Quick Path Interconnect (QPI). This provides a high speed interface both between the CPUs themselves and out to the rest of the system. Each QPI link delivers no less than 32GB/s and each CPU sports two links. For workloads where data must be shared between CPU sockets, it's an enormous boon

Intel To Launch Nehalem Server Processor

Intel (NSDQ: INTC) is scheduled to launch on Monday Nehalem EP, the server version of the chipmaker's next-generation microarchitecture. But while the new features within Nehalem EP coincide with the direction of computing, corporate adoption is likely to be hampered by the economic downturn.
Intel is expected to introduce Nehalem EP processors and highlight the systems of major computer makers, such as Hewlett-Packard, Dell, and IBM, at its corporate headquarters in Santa Clara, Calif.
Analysts agree that Nehalem EP, with its integrated memory controller for better performance, is poised to deliver more power to meet the increasing demand of virtualization in the data center. In addition, Nehalem-based processors should help with the push toward cloud computing, which typically refers to the running of applications in an Internet server or downloading the software from the Internet each time it's used. Google Apps is an example of business applications delivered via cloud computing.
Initially, Nehalem EP processors will primarily be available with four cores. Intel plans to introduce a six-core Nehalem processor and an eight-core design, called Nehalem EX, by the end of the year.
But sales of the new Intel products are expected to be slow this year, as companies riding out the economic recession cut IT spending and delay projects. "It's pretty clear that companies are going to be very conservative with the money they spend this year, and that's going to slowdown adoption," John Spooner, analyst for Technology Business Research, told InformationWeek.
Where companies typically replaced old servers every three years, many businesses are likely to stretch out that replacement cycle by an additional year or two, Spooner said.
Indeed, worldwide server shipments and revenue fell 11.7% and 15.1%, respectively, in the fourth quarter of last year from the same period a year ago, according to IDC. The downward trend in spending is expected to continue this year.
If companies delay purchases of Nehalem EP-based servers this year, then it's unclear how customers will react to the launch next year of Nehalem processors based on Intel's 32-nanometer manufacturing process, which will produce faster, more energy efficient chips than the current batch of 45-nm processors. Companies could skip this year's Nehalem EP systems in favor of those based on the 32-nm variant, code-named Westmere, next year

Fresh Crop Of Intel Nehalem Servers On Way

Dell, HP, IBM, Lenovo, and Sun are among the vendors launching products at the same time Intel officially introduces its latest Xeon server chips.

Major computer makers are making sure there's no shortage of servers powered by Intel (NSDQ: INTC)'s latest Nehalem processors, but despite the abundance, there's not a huge difference in hardware and buyers should pay closer attention to the software inside the systems.
Dell (Dell), Hewlett-Packard (NYSE: HPQ), IBM (NYSE: IBM), Lenovo, and Sun Microsystems (NSDQ: JAVA) are among the vendors launching products at the same time Intel officially introduces Nehalem EP server processors, which are based on a new microarchitecture with an integrated memory controller for better performance. The Nehalem variant, known as Core i7, for high-end desktops and workstations was launched late last year.

Over the last week, vendors have been vying for media attention with embargoed news conferences and press releases to make the case for their respective products. But despite the many benchmarks vendors are showing to prove faster performance than competitors' systems, there's actually not a whole lot of difference on the hardware level, analysts say.
"We're getting down to literally microseconds of difference in terms of performance and power," Andi Mann, analyst for Enterprise Management Associates, told InformationWeek.
Nevertheless, all the vendors are claiming significant performance boosts over previous generations of servers, which make the latest products better for virtualization. This means IT managers can reduce the overall number of servers in a data center by running more business applications in a single system.
Where vendors have their best chance to differentiate their new products is in the software that ships with their servers, particularly the system management applications. Tools that ease the task of provisioning software across servers, provide notification when there's a problem, monitor power usage, and offer other management features are more likely to attract the attention of potential buyers.
That fact is not lost on the vendors. Hewlett-Packard on Monday introduced the ProLiant G6 line of 11 Nehalem EP-powered tower, rack, and blade servers, the largest ProLiant rollout ever.
Among the key system management features is energy efficiency. HP has added sensors across the line that automatically adjusts system components such as fans, memory, and input/output processing to the workload. In addition, IT managers can cap the power drawn by individual servers.

Intel's Nehalem Server Chips Seen Aiding Virtualization

Intel on Monday introduced 17 Nehalem EP-based quad-core processors, officially launching its next-generation microarchitecture that one analyst said fixes the most obvious technical weaknesses in Intel's previous server platform.
Patrick Gelsinger, senior VP and general manager of Intel's Digital Enterprise Group, told a news conference at the company's Santa Clara, Calif., headquarters that the new processors represented the biggest platform advancement for Intel in more than 10 years.
"This is as significant and as transformational as the Pentium Pro was in its day," Gelsinger said.
Intel believes Nehalem EP, officially called the Xeon 3500 and 5500 series, provides the kind of technology organizations will need as they deploy more virtualization into their data centers and move toward cloud computing, which typically refers to the running of applications in an Internet server or downloading the software from the Internet each time it is used. Google Apps is an example of business applications delivered via cloud computing.
Virtualization and cloud computing are seen by the industry as big drivers for future server demand. Cloud computing is still in its infancy, and the former is starting to take off. Analyst firm Gartner predicts virtualization, or the use of virtual machines to run multiple business applications on different operating systems in the same computer server, will be the No. 1 technology initiative this year.
With this in mind, Intel has dramatically boosted the performance of its microarchitecture by introducing an integrated memory controller on the same piece of silicon as the processor, thereby eliminating the front-side bus that analysts said was a bottleneck in Intel's previous server platform. Intel rival Advanced Micro Devices has used an IMC in its Opteron processors since 2003.
In getting rid of the FSB, Intel has eliminated its "traditional area of weakness," Nathan Brookwood, analyst for Insight 64, said. "This is the last of the low-hanging fruit."
Going forward, Intel will have to work harder at refining the technology it has to squeeze out more power and energy efficiency. "There's no reason to shut down the patent office," Brookwood said.
While Nehalem EP gives Intel a performance lead over AMD, the gain is only expected to last for six months or so. AMD is sure to catch up with its future products. "We're in a leapfrog kind of mode," Brookwood said.
"This is as significant and as transformational as the Pentium Pro was in its day," Gelsinger said.
Intel believes Nehalem EP, officially called the Xeon 3500 and 5500 series, provides the kind of technology organizations will need as they deploy more virtualization into their data centers and move toward cloud computing, which typically refers to the running of applications in an Internet server or downloading the software from the Internet each time it is used. Google Apps is an example of business applications delivered via cloud computing.
Virtualization and cloud computing are seen by the industry as big drivers for future server demand. Cloud computing is still in its infancy, and the former is starting to take off. Analyst firm Gartner predicts virtualization, or the use of virtual machines to run multiple business applications on different operating systems in the same computer server, will be the No. 1 technology initiative this year.
With this in mind, Intel has dramatically boosted the performance of its microarchitecture by introducing an integrated memory controller on the same piece of silicon as the processor, thereby eliminating the front-side bus that analysts said was a bottleneck in Intel's previous server platform. Intel rival Advanced Micro Devices has used an IMC in its Opteron processors since 2003. In getting rid of the FSB, Intel has eliminated its "traditional area of weakness," Nathan Brookwood, analyst for Insight 64, said. "This is the last of the low-hanging fruit."

Dell Offers HD Screen With Mini 10 Netbook

Dell (Dell) on Tuesday said it's offering the option of a high-definition display with the Inspiron Mini 10 netbook.
The Internet's influence is much in evidence in Poland's capital city – despite the best efforts of the French to bollocks it up.
The mini-laptop comes standard with a 10-inch 1,024- by 576-pixel display. However, Dell recently added the option of a 1,366- by 768-screen for an additional $35.
The optional resolution is the equivalent of 720p, which is shorthand for a high-definition video mode. The number 720 stands for 720 pixels of vertical resolution, while the p stands for progressive scan or noninterlaced. The resolution has a widescreen aspect ratio of 16:9.
Other options available with the Mini 10 include a faster Intel (NSDQ: INTC) Atom processor for an additional $50. Starting price for the mini-notebook is $399.
Netbooks are a category of PC defined as having screens 10 inches or less and costing less than $500, with many models selling for as little as $300. The systems have limited performance, but are adequate for Web browsing and e-mail.
The mini-notebooks have proven very popular in the economic recession. Global sales will grow nearly 80% this year to capture 8% of total PC sales, according to the latest Gartner forecast.
Last year, netbook shipments reached 10 million units, with half of the units sold in the fourth quarter, according to IDC. The devices accounted for 7% of overall mobile PC shipments.

IBM-Sun talks may extend beyond next week – source

IBM's talks to acquire Sun Microsystems Inc are continuing and may extend beyond next week, according to a person with knowledge of the matter.
IBM is still examining Sun's business as part of its due diligence process, said the source, who was not authorized to speak about the talks and therefore requested anonymity.Neither IBM nor Sun has issued any statement to say they are in talks, although sources said last week that the two sides are negotiating a merger that would bolster IBM's high-end server and software business.The Wall Street Journal reported on March 18 that IBM could pay as much as US$8 billion for Sun, amounting to a 100 percent premium for the high-end server computer maker. If a deal is sealed, it would be IBM's largest acquisition.The source said on Thursday that IBM's due diligence process, or examination of Sun's business, was necessary considering Sun's size and complexity.An IBM spokesman declined to comment, and Sun was not immediately available.Some analysts have said Sun would bolster IBM's position against rivals like Hewlett-Packard and Cisco Systems, both of which have been acquiring smaller, niche technology firms to broaden their product and service offerings.Sun shares were up 1.5 percent at US$7.97 in early afternoon trading, shedding some of their earlier gains to US$8.23. IBM shares were up 0.7 percent at US$98.64.